Floor-ladder framework for human face beautification

Abstract

In this paper, we propose a Floor-Ladder Framework (FLN) based on age evolution rules to generate beautified human faces. Beside the shape of faces, younger faces achieve more attractiveness. Thus we process the beautiful face by applying the reversed aging rules. Inspired by the layered optimization methods, the FLN adopts three floors and each floor contains two ladders: the Single Layer Older Neural Network (SLONN) and the extended Skull Model. The Peak Shift algorithm is designed to train the SLONN aiming to capture the reversed aging rules of the face skin. Due to the growth rules of the face shape, we extended the Skull Model by adding Marquardt Mask. Given the input portrait, our algorithm effectively produces a beautified human face without losing personal features.